DADL: A Declarative Description Language for Enterprise Tool Libraries in LLM Agent Systems

2026-05-08T08:51:59Z9fd64c66f856dfd80033ecd80203bc6e23ab7f935b28af1bdb2f4b990c2abcb6
API integrationCI/build failure analysisDADLLLM agentsagent workflowsaudit automationcode generationdata poisoning riskdocument intelligenceenterprise tool catalogethics governancefull-text indexingknowledge graphproactive coding assistantsrepository context filessecurity implicationssoftware supply chaintraining-data quality

What happened

Batch of CS-SE arXiv papers on LLM agents, developer tooling, and AI-assisted software engineering. Highlights: DADL (Dunkel API Description Language) proposes a declarative YAML for REST integrations and a shared runtime to create versioned Enterprise Tool Libraries (public registry ~1,833 tool defs; Code Mode reduces tool-advertisement context from ≈142k to ≈1k tokens). Other papers: AI-based document intelligence for population-level audit testing; PARNESS, an open-source framework for dynamic agentic research with full-text indexing and a knowledge-graph index; a systematic review mapping

Why it matters

A reviewed impact interpretation has not been published for this record.

Evidence and limitations

Source ID
arxiv_cs_se
Record identifier
9fd64c66f856dfd80033ecd80203bc6e23ab7f935b28af1bdb2f4b990c2abcb6
Enrichment time
2026-05-08T08:51:59Z
AI-assisted enrichment
Yes

This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.